JPH06131000A - Fundamental period encoding device - Google Patents

Abstract

PURPOSE:To attain a low bit rate and to attain high performance in the case of differentially encoding a fundamental period parameter. CONSTITUTION:This device is provided with a fundamental period preliminary selection part 40 for outputting the pair of candidate fundamental period corresponding to the respective sub frames of a relevant frame based on an input voice signal, and fundamental period main selection part 60 for deciding a fundamental period by using the input voice signal of the relevant sub frame, the spectrum envelop parameter of the relevant sub frame and parameter representing excitating sound source decided in the past from among adjacent fundamental period parameters in a range previously decided for the fundamental period parameter corresponding to the relevant sub frame in the pair of candidate fundamental period outputted from the fundamental period preliminary selection part 40.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to low bit rate coding of a speech signal, and more particularly to a coding device for a speech fundamental period parameter.

[0002]

2. Description of the Related Art CELP (Code-Excited Linear), which is a combination of a linear predictive coding method and vector quantization, is used as a method for efficiently coding a voice signal.
Prediction) coding method. This method is
It consists of a linear predictor that represents the spectrum envelope of the signal, a codebook that represents the fundamental period of speech, a source codebook that represents the excitation signal, and a gain codebook that performs amplitude control. In CELP, the linear predictor is usually obtained in units of frames divided at regular intervals, and the other components are usually determined for each subframe obtained by further dividing the frame.

As a method of encoding the basic period determined by the adaptive codebook section for each subframe at a low pit rate, M. Yong and A. Reference (1) by Gersho: "Efficient encoding o"
f the long-termpredictor
in vector excitation code
res, ", and Campbell et al. (2):" An Expander Error
-Protected 4800 bps CELP
Coder (U.S. Federal Standard)
d 48000 bps Voice Coder). The former encodes the basic period for each subframe with the average period over the entire frame and its variation,
It is possible to reduce the sound quality and reduce the bit rate. On the other hand, the latter is a method in which the fundamental period found in odd subframes is encoded by the difference from even subframes.
It also has the effect of reducing the amount of calculation required for basic period search. On the other hand, in the method of extracting the basic cycle only from the information of the input speech signal instead of the basic cycle determined by the adaptive codebook unit, a configuration in which the basic cycle is differentially encoded is disclosed in, for example, Document (3): JP-A-2-216200. It is proposed in "Speech Coding Device and Speech Coding Device".

[0004]

First, the above-mentioned document (1) in which the fundamental period is encoded as an average value and a variation from the average value.
The conventional method of (1) has a problem that it is not efficient as compared with the successive difference method of encoding the difference from the basic period defined in the past subframe. For example, when the basic period fluctuates {8, 10, 12, 14, 16} for each subframe, in the conventional method of the above-mentioned document (1), the average 12 is represented by 4 bits, and the fluctuation from the average. 7 (= 2 + 1
+ 1 + 1 + 2) bits require a total of 11 bits. But,
When encoding sequentially, 7 (= 3 + 1 + 1 + 1 +
1) It becomes a bit and the bit rate can be further reduced.

On the other hand, in the conventional method of the above-mentioned document (2) in which quantization is performed by the difference between the even subframes and the odd subframes, when the basic period of the odd subframe is obtained, the basic period determined by the past even subframe is set. Since the presupposed sequential decision is adopted, the means for encoding the fundamental period is not designed to determine the optimal set of fundamental periods for the entire frame, and the characteristics are large compared to the case without differential encoding. There is a problem of deterioration. In addition, since the number of bits assigned to odd-numbered subframes is constant, there is a drawback in that when the audio signal varies within the frame, the variation cannot be followed. Furthermore, in order to apply the latter to a lower bit rate, it is necessary to reduce the number of bits in odd subframes, which causes a problem of deterioration in characteristics.

Further, the differential encoding of the fundamental period in the conventional system which does not have the fundamental period part for minimizing the error between the input voice signal and the reproduced voice signal as in the literature (3) is a simple time series of the fundamental period parameter. This is when viewed as a parameter. Since the distance between the reproduced voice signal and the input voice signal is ignored, there is a problem that a high quality reproduced voice signal cannot be obtained.

It is an object of the present invention to provide a basic period coder that encodes a basic period at a low bit rate so as to be optimal for the entire frame, and to improve the quality of a speech coding formula. is there.

[0008]

A fundamental period coder according to a first aspect of the present invention uses a discrete speech signal with a parameter representing a spectrum envelope, a parameter representing a fundamental period, a parameter representing an excitation source, and a parameter for performing amplitude control. In the speech coding method to express, when a discrete speech signal divided into frames at fixed intervals is input and the basic period parameter is coded for each subframe in which the frame is dispersed into a plurality of sections, In a basic period coder that encodes some of the period parameters by the difference with the basic period parameter of the subframe determined in the past, in the candidate basic period corresponding to each subframe of the frame based on the input speech signal. And a basic period preliminary selection unit that outputs a set, and the subframe of the candidate basic period set output from the basic period preliminary selection unit. From the neighboring fundamental period parameter within a predetermined range with respect to the fundamental period parameter corresponding to the frame, using the input speech signal of the subframe, the spectrum envelope parameter of the subframe, and the parameter representing the excitation source defined in the past. A basic period main selecting unit that determines a period, wherein the basic period preliminary selecting unit uses the input speech signal in a process of obtaining a candidate basic period of a subframe over the entire frame in the basic period preliminary selecting unit. A plurality of basic period parameter candidates for each subframe and a prediction gain associated with the candidate are defined, and the difference between the basic parameters between adjacent subframes based on each of the determined basic period parameter candidates is predetermined. A basic period parameter set within the specified range, and And extracting based on the candidate fundamental period prediction gain associated with them from the set of synchronization parameter.

The basic period coding apparatus of the second invention is the first invention.
In the process of the invention, the basic cycle preliminary selecting unit selects a subframe that gives a maximum prediction gain by using the input speech signal in the process of obtaining a candidate basic cycle parameter of the subframe over the entire frame in the basic cycle preliminary selecting unit. A basic period parameter candidate based on the input speech signal between the determined subframes, and a prediction gain associated with the candidate, and adjacent subframes based on each of the determined basic period parameter candidates. A basic period parameter set in which the difference between the basic parameters is within a predetermined range, and a candidate basic period is set from the set of basic period parameters over the entire configured frame based on the prediction gains associated with them. It is characterized by extracting.

A basic period coding apparatus according to a third invention is the first invention.
In the invention, the basic cycle main selection unit constitutes a set of neighboring basic cycle parameters within a predetermined range from the basic cycle determined in the past subframe, It is characterized in that the fundamental period is determined by using the input signal of the subframe, the spectrum envelope parameter of the subframe, and the parameter representing the excitation sound source determined in the past.

A basic period coding apparatus according to a fourth invention is the first invention.
In the invention, the basic cycle preliminary selecting section outputs a plurality of candidate basic cycles.

According to a fifth aspect of the present invention, there is provided the basic period coding device according to the first aspect.
The invention is characterized in that the number of bits assigned to the basic period to be differentially encoded is different for each subframe.

According to a sixth aspect of the present invention, there is provided a basic period coding device which comprises:
In the invention, a mode discriminating unit that divides the frame or the subframe into a predetermined class based on a prediction gain for each subframe calculated from the input speech signal is provided, and output from the mode discriminating unit. It is characterized in that the number of bits allocated to the basic period to be differentially encoded is made different for each subframe based on the class information.

A basic period encoding apparatus according to a seventh aspect of the invention is the first aspect.
In the invention of, the basic cycle preliminary selection unit, when extracting a candidate basic cycle from a set of a plurality of lags configured so that the difference in basic parameters between adjacent subframes is within a predetermined range, It is characterized in that the set of basic parameters is narrowed down to a smaller number of sets by using a prediction gain associated with each lag, and the highest basic period is extracted from the set of narrowed down basic parameters.

The basic period coding apparatus according to the eighth invention is the first invention.
In the invention, in the process of the basic cycle preliminary selecting section configuring a candidate basic cycle of a subframe in the basic cycle preliminary selecting section, a prediction gain determined by the frame, the frame, and an input speech signal of the next frame is set. The range of the basic parameter difference between adjacent subframes is a predetermined range over a plurality of frames, that is, a part of the subframe of the previous frame, a subframe of the frame and a part of the subframe of the next frame. And a candidate basic period is extracted from the set of basic period parameters formed over the plurality of frames based on the prediction gains associated with them.

A basic period encoding apparatus according to the ninth invention is the first invention.
In the invention, the basic period preliminary selection unit comprises a plurality of sets of basic period parameters within the predetermined range from the basic parameter candidates of adjacent subframes, and the plurality of sets of basic period parameters are configured. When determining a candidate basic period from the set, the basic period is extracted using a basic period parameter that gives a prediction gain exceeding a predetermined threshold value among the elements of the set of basic parameters.

[0017]

Now, the input signal of the i-th frame is {s
_i (n)} and the subframe length is L. At this time, for each subframe nsub = 0, ..., N−1,

[0018]

It is assumed that has been calculated. Here, t is a term representing a time lag (or a cycle of the input signal), and the prediction gain for each t is

[0020]

Is calculated as still,

[0022]

It is

The basic cycle preliminary selection section in the first aspect of the invention executes the processing by the following actions. Using the input speech signal s (n), the prediction gain for the lag t _nsub of each sub-frame is determined using the equation (3) or (4). Then in each subframe

[0025]

The following relationship is satisfied. Hereinafter, obtaining a set of lags {l _i } satisfying such a relationship will be referred to as tracking. Also, here, in order to obtain {l _i },

[0027]

[0028] It can be processed by repeatedly using lags of subframes located in the future. Finally, satisfy the above equation {l
_i } of

[0029]

Alternatively,

[0031]

The lag set l _imax that maximizes the cumulative prediction gain is selected and sent to the basic period main selection unit.
In the basic period main selection unit according to the first aspect of the present invention, the linear prediction coefficient {α (i)} representing the spectrum envelope of the voice and the past excitation sound source {v (n), n = 0 ,. ． ． , L _c −1} and the sub-frame of l _imax output from the basic preliminary selection section.

[0033]

[0034]

[0035]

Here,

[0037]

[0038] Differential encoding.

In the above equation, an integer value is assumed as the lag value t. Kroon an
d B. S. Reference by Atal (4): "Pitch
Predictors With High Te
general Resolution, "Proc.I
EEE Inc. Conf. on. Acoustic
s, Speech and Signal Proce
ssing, pp. 661-664 1990. It is also possible to use fractional values such as those described in "." In this case, the fractional values are indexed to integers and differential encoding is applied to the index difference. Therefore, it is possible to adopt a configuration in which only an integer lag is searched when calculating a basic period path candidate and a decimal lag is searched when extracting a candidate basic period.

In the basic cycle preliminary selecting section in the second invention, first, similarly to the first invention, the prediction gain for each lag in each subframe is calculated. Then in the subframe

[0041]

The following relation is satisfied. And finally, of {l _i } satisfying the above equation,

[0043]

The lag set l _imax for maximizing the cumulative prediction gain is selected and sent to the basic cycle main selection section. The second invention has an effect of giving the same characteristics with a smaller amount of calculation as compared with the first invention.

In the basic period main selection section in the third invention,

[0046]

[0047] There is.

[0048]

[0049] it can

In a fourth aspect of the present invention, a basic cycle preliminary selecting section for outputting a plurality of sets of lags {l _k } (k> 1) to be sent to the basic cycle preliminary selecting section, and a plurality of basic cycle preliminary selecting sections for outputting the basic cycle preliminary selecting section. The present invention has a main selection unit that performs the selection shown in the above formula (9) for lags near a predetermined range of lags, and compared with the above invention narrowed down to a single candidate. This has the effect that a near-optimal basic period parameter can be obtained.

A fifth aspect of the present invention makes the number of bits required for differential encoding different for each subframe.
c and the WD are functions of subframes. For example, the basic cycle is {8, 3, 7, 3, for each subframe.
When encoded with 3} bits, WD is WD (nsub)
= {256,8,128,8,8}, and accordingly W _c also becomes W _c (nsub) = {8,8,16,8,8}. In addition, when changing the value of W _c , the track Can be

A sixth aspect of the present invention makes the number of bits required for differential encoding for each subframe different according to the characteristics of the frame or the subframe. The characteristics of the frame or the subframe are as follows. Information such as the geometric mean of the prediction gain for each subframe or the geometric mean of the RMS for each subframe can be used.
Classifying is performed in a predetermined manner, and the number of bits for the differential code and the W _{c and the} WD are also assigned subframes according to the class. Geometric mean

[0053]

The frame is classified into the following predetermined classes according to the value of.

0 ≤ Pgn <0.25 CLASS0 0 0.25 ≤ Pgn <0.5 CLASS1 0.5 ≤ Pgn <0.75 CLASS2 0.75 ≤ Pgn <CLASS3 According to this classification, the number of bits allocated to differential encoding Are switched as follows.

WD (nsub = 0, ..., 3) {0,0,0,0,0} for CLASS0 WD (nsub = 0, ..., 3) {8,3,7,3,3 } For CLASS1 WD (nsub = 0, ..., 3) {8,3,5,4,4} for CLASS2 WD (nsub = 0, ..., 3) {8,4,4,4,4 } For CLASS3 It is also possible to change W _c (nsub) according to the allocation of this bit. According to the present invention, it becomes possible to encode the basic cycle according to the characteristics of the input voice of the frame.

A seventh aspect of the invention is a predetermined value from the maximum cumulative error PG (imax), not the cumulative error of the prediction gain, when obtaining l _imax described in the first and second _aspects of the invention. Out of the set of lags that give a cumulative error within, the one with the smallest lag is output as the basic period preliminary selection unit. The present invention can prevent an error in extracting a double pitch.

An eighth aspect of the invention not only applies the procedure of the basic preliminary selection section to the audio signal of the frame, but also obtains l _imax using the audio signal of the subframe of the previous frame or the next frame. . For example, an expression for obtaining the lag tracking l _i

[0059]

Then, the formula for calculating the accumulation

[0061]

In, the value of nsub is nsub =
{-1, 0, 1 ,. ． ． , N-1, N}. At this time, nsub = −1 indicates that the prediction gain of the N−1th subframe of the previous frame is used, and nsub = n
sub = N indicates that the prediction gain of the 0th subframe of the next frame is used. The present invention has the effect of reducing discontinuity between frames.

The ninth invention is the same as the first invention described above.
_{When imax} is calculated, the cumulative prediction gain cumulative value is calculated only for the prediction gain exceeding the predetermined prediction gain threshold gn _th . That is,

[0064]

According to the present invention, in the case where the periodic structure is changed in a frame, the subframe having more distinct periodicity can be weighted to extract the candidate basic period.

Up to this point, the third is the cumulative prediction gain.
The formula or the formula 4 was used, but the arithmetic mean of the prediction gains for each subframe is also used.

[0067]

Anything can be used as long as it indicates the degree of periodicity of the input audio signal.

This completes the description of the operation of the present invention.

[0070]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing an embodiment of the basic period coding apparatus of the first invention. The audio signal input from the input terminal 10 is stored in the frame buffer 20. The linear predictive analysis unit 30 supplies the input audio signal of the frame from the frame buffer 20, and performs linear predictive analysis well known in the literature (4): Saito, Nakata, Basics of audio information processing, for example, for LSP. Such a linear prediction parameter 1z {lsp _k (i)} is obtained and output to the LPC quantizing unit 50, the LPC memory 31, and the LPC interpolating unit 32. The LPC quantizer 50 applies the quantization method described in the above-mentioned document (4) to the LSP parameter {lsp _k (i)} supplied from the LPC analysis unit 30 to determine the quantized LSP { I asked the qlsp _k (i)},
It outputs to the quantized LPC interpolation unit 52 and the quantized LPC memory 51. Further, the index information obtained at the time of quantization is sent to the output terminal 14. In the LPC interpolator 32,
LSP of previous frame in PC memory 31 {lsp
_k−1 (i)} and the LSP {lsp _k (i)} of the frame supplied from the LPC analysis unit 30 are linearly interpolated to obtain the linear prediction coefficient {α of the subframe nsub.
_nsub (i)} is obtained, and the basic cycle preliminary selection unit 4
0, the fundamental period main selection unit 60, the excitation source search unit 70, and the gain search unit 90. In the quantized LPC interpolator 52, the quantized LSP {qlsp _k−1 (i)} of the previous frame supplied from the quantized LPC memory 51 and the quantized L of the frame supplied by the LPC quantizer 50.
SP {qlsp _k (i). } Is linearly interpolated and the linear prediction coefficient {β _{ns ub} (i)} of the subframe nsub is output to the basic period main selection unit 60, the excitation source search unit 70, and the gain search unit 90.

Here, the structure of the basic cycle preliminary selecting section 40 will be described with reference to FIG. First, in the weighting unit 41,
The input audio signal supplied from the frame buffer 20 is set to L
The linear prediction coefficient {α supplied from the PC interpolation unit 32
_nsub (i)} is used for conversion as follows.

[0072]

Here, v (i) and u (i) are arbitrary coefficients, for example, v (i) = 0.9 ⁱ , u (i) = 0.4 ^i.
And so on. Next, the input speech signal {s _w (n)} weighted by the prediction gain calculation unit 42 is {s (n)}.
The prediction gain is calculated by the procedure described in the above operation. That is,

[0074]

In the basic cycle path candidate construction unit 43, first,
Cycle tma that gives the maximum value prediction gain : "Techniques for improvin
g the performance of CELP
type speech coderes, "Pro
c. IEEE Inc. Conf. on. Accoust
ics, Speech and Signal Pro
cessing, pp. 205-2081991. From the divisors and multiples of tmax _nsub , as described in It is also possible to change it according to the state of the frame and the magnitude of the subframe maximum prediction gain. For example, when the maximum prediction gain of the previous frame is smaller than 0.3, T _h
= 0.5, 0.3, it is possible to set T _h = 0.75. _0.3, T h = _0.75, and the like otherwize. Each subframe obtained in this way

[0076]

A set of lags satisfying the relation is obtained. The {l _i } thus obtained is output to the candidate basic period extraction unit 44. In the candidate basic cycle extraction unit 44,
From {l _i } supplied from the basic period path candidate configuration unit, a set of lags l for maximizing the cumulative prediction gain of the following equation:
_imax is selected and output to the basic period main selection unit 60.

[0078]

This completes the explanation of the basic cycle preliminary selection section 40. Returning to FIG. 1, in the basic cycle main selection 60, the input audio signal {s (n-nsub × L)} of the subframe supplied from the frame buffer 20 and the excitation supplied from the past excitation sound source memory 80. Sound source signal {v (n)}
And the LPC coefficient {α _nsub (i)} of the subframe supplied from the LPC interpolator 32, and the quantized LPC coefficient {β _nsub (i)} of the subframe supplied from the quantized LPC interpolator. Basic cycle preselection using

[0080]

[0081]

[0082]

Where

[0084]

[0085] The source {e (n)} is output to the excitation sound source searching unit 70 and the gain searching unit 90. The excitation sound source search unit 70 uses the sound source code vector c in the sound source code book C stored in advance.
_{From i} (n), the one that minimizes the following equation c _imax (n)
Select.

[0086]

The selected sound source code vector is sent to the gain search section 90, and its index imax is the output terminal 12
Is output to. Also, the newly determined excitation sound source signal: g
_e e (n) + g _v c _imax (n) is the excitation sound source memory 8
Sent to 0. The gain search unit 90 uses the gain code vector G
_From ( _e (k), g _v (k)), which minimizes the following equation (g
_e (kmax), g _v (kmax)).

[0088]

The obtained gain code vector is output to the excitation sound source memory 80, and its index kmax is output to the output terminal 13. In the excitation sound source memory 80, g _e (k
max) e (n) + g _v (kmax) c _imax (n)
Is stored as a past excitation source. The excitation sound source memory 8
0 is updated as follows.

V (n−L) = v (n), n = L ,. ． ． , L _c −1 (43) v (L _c −L + n) = g _e (kmax) e (n) + g _v (kmax) c _{im ax} (n), n = 0 ,. ． ． , L-1 (44) FIG. 3 is one of the embodiments of the basic period preliminary selection section in the basic period encoding device of the second invention. The basic preliminary selection section in this embodiment operates as follows. The prediction gain calculation unit 42 uses the same procedure as described in the above embodiment to predict each subframe. Subframe giving the maximum prediction gain of the prediction gains supplied from the calculation unit 42 Force In the basic cycle path candidate configuration unit 243, first, the maximum prediction gain extraction unit 24

[0091]

[0092] It The {l _i } determined in this way is the candidate basic period extraction unit 4
4 is output. In the candidate basic cycle extraction unit 44, from among {l _i } supplied from the basic cycle path candidate configuration unit, a set of lags l _imax that maximizes the cumulative prediction gain of the following equation.
Is output to the basic period main selection unit 60.

[0093]

FIG. 4 shows an embodiment of the basic period main selection section in the third invention. First, the operation in the first subframe will be described. In the candidate lag constructing unit 362, the tracking l supplied from the basic cycle preliminary selecting unit 40 of FIG.
Pair from _imax to the first subframe

[0095]

[0096] To be In the distance calculation unit 361, the lag set supplied from the candidate lag configuration unit 362

[0097]

The calculated D (o) is sent to the basic period lag selector 363. In the basic period lag selector, the above {D
From (o)}, find omax that gives the maximum value,
Candidate lag It is output to the output terminal 11 of FIG. Excitation sound source configuration unit 3
In 64, basic cycle lag selection And from

[0099]

And outputs {g _e e (n)} to the excitation source search unit 70 of FIG. Next, the operation after the second subframe will be described. Only the processing of the candidate lag constructing unit 362 is essentially different from the first subframe. In the candidate lag configuration unit 362, the basic cycle Neighborhood

[0101]

[0102] } And the corresponding {e (n), n = 0 ,. ． ． ，
L × N−1} is decided.

FIGS. 5A and 5B show an embodiment of the basic cycle preliminary selecting section and the basic cycle main selecting section in the fourth invention. In the candidate cycle extraction unit 444 of the basic cycle extraction unit 440 in this embodiment, the cumulative prediction gain is calculated from the set of lags {l _i } supplied from the basic path candidate configuration unit 43.

[0104]

[0105] M-1} is selected and output to the basic period main selection unit 460.
In the candidate lag configuration unit 462 of the basic cycle main selection unit 460,
A plurality of rasters supplied from the basic cycle preliminary selection unit 440

[0106]

[0107] Supplied from part 462

[0108]

[0109] It is output to the vibration source configuration unit 364.

FIG. 6 shows an embodiment of the basic cycle preliminary selecting section and the basic cycle main selecting section in the fifth invention. In the basic cycle path candidate configuration unit 543 of the basic cycle preliminary selection unit 540,
From the candidate range table 501, the tracking search range {W _c (ns

[0111]

The following relation is satisfied. The lag set l _i tracked in this way is stored in the candidate basic period extraction unit 44.
Is output to. The candidate basic cycle extraction unit 44 selects a lag set l _imax that maximizes the cumulative prediction gain of the following expression from the {l _i } and sends it to the candidate lag configuration unit 562 of the basic cycle main selection unit 560. Output.

[0113]

[0114] At the time of generation, the range {WD (nsub)} for each subframe is input from the candidate range table 501,

[0115]

[0116] is satisfied. Distance calculator 36
1 is supplied from the candidate lag component 562

[0117]

Calculate Other processes are the same as those described in the above embodiments.

FIG. 7 shows an embodiment of the basic cycle preliminary selecting section and the basic cycle main selecting section in the sixth invention. In the frame mode identification unit 601, first, from the prediction gain output from the prediction gain calculation unit 42 of the basic cycle preliminary selection unit 640, the maximum of each subframe is calculated.

[0120]

Alternatively, the arithmetic mean

[0122]

Is calculated, and finally the class of the frame is judged by the value. For example, the following table is used for class classification.

0 ≦ Pgn <0.25 CLASS0 0.25 ≦ Pgn <0.5 CLASS1 0.5 ≦ Pgn <0.75 CLASS2 0.75 ≦ Pgn <CLASS3 In the above classification, rm in each subframe.
The s value can be used together, and (0 ≤ Pgn <0.25) ∩ (0 ≤ rms <500) CLASS0 (0.25 ≤ Pgn <0.5) ∩ (500 ≤ rms <900) CLASS1 (0. 5 ≤ Pgn <0.75) ∩ (800 ≤ rms <1200) CLASS2 (0.75 ≤ Pgn) ∩ (1000 ≤ rms) CLASS3. Class index cld
ex is an output terminal 602, a basic cycle path candidate configuration unit 643 of the basic cycle preliminary selection unit 640, and a basic cycle main selection unit 660.
Is output to the candidate lag constructing unit 662. In the basic cycle path candidate configuration unit 643, the tracking gain range W _c (ns) in the subframe determined by the prediction gain of each subframe supplied from the prediction gain calculation unit 42 and the class index supplied from the frame mode identification unit.
ub) and tracking {l _i } are output. Here, it is assumed that the tracking search range W _c is determined as follows.

W _c (nsub = 0 ,,,, 3) {0,0,0,0,0} for CLASS0 W _c (nsub = 0 ,,,, 3) {256,8,128,8,8 } For CLASS1 W _c (nsub = 0 ,,,, 3) {256,8,32,8,8} for CLASS2 W _c (nsub = 0 ,,,, 3) {256,16,16,16,16 } For CLASS3

[0126]

The following relationship is satisfied. The tracking set l _i determined in this way is output to the candidate basic period extraction unit 44. The candidate basic cycle extraction unit 44 calculates the cumulative prediction gain of the following expression from among {l _i }

[0128]

The _value l _imax that _maximizes is obtained and output to the candidate lag constructing unit 662 of the basic period main selecting unit 660. The candidate lag constructing unit 662 selects the lag to be output to the distance calculating unit 361. For example, WD (nsub = 0, ..., 3) {0,0,0,0,0} for CLASS0 WD (nsub = 0, ..., 3) {128 , 4,64,4,4} for CLASS1 WD (nsub = 0, ..., 3) {128,4,16,4,4} for CLASS2 WD (nsub = 0, ..., 3) {128 , WD (nsub)} determined by a table for FOR CLASS3 and the basic cycle preliminary selecting section 640.

[0130]

[0131]

[0132]

Calculate Other processes are the same as those described in the above embodiments.

FIG. 8 shows an embodiment of the basic cycle preliminary selecting section in the seventh invention. In the candidate basic cycle extraction unit 744 of the basic cycle preliminary selection unit 740, first, the fundamental frequency path candidate configuration unit Cumulative prediction gain of

[0135]

Is calculated and then a set of lags within a predetermined range from its maximum value PG _max.

[0137]

[0138] Is output to the basic period main selection unit 60. In the above equation, T _th
Is an arbitrary threshold and can be taken as 0.9, for example.

FIG. 9 shows an eighth embodiment of the basic cycle preliminary selecting section. In this embodiment, the input signal {s (n)} supplied from the input terminal 10 is the 2-frame buffer 820.
Is stored in. The input audio signal of the frame is supplied from the 2-frame buffer 820 to the LPC analysis section 30, and the input audio signals of the two frames of the frame and the next frame are supplied to the weighting section 841 of the basic cycle preliminary selection section 840.
In the weighting unit 841, the linear prediction coefficient {α supplied from the LPC interpolation unit 32 is applied to the input audio signal of (1 + 1 / N) frames supplied from the two-frame buffer 820.
_nsub (i)} is used for conversion as follows.

[0140]

The linear prediction coefficients of nsub = N-1 and nsub = N in the above equation are calculated as the same. In the prediction gain calculation unit 842, {s supplied from the weighting unit 841
Prediction gain for (N + 1) subframes from _w (n)}

[0142]

Is calculated based on the third or fourth equation described in the above operation. In the basic cycle path candidate configuration unit 843, first, nsub = {0 ,. ． ． , N} for each subframe Previous frame supplied from the cycle memory 845 and determined by the candidate basic cycle extraction unit Is output. The candidate basic cycle extraction unit 844 selects a lag set l _imax that maximizes the cumulative prediction gain of the following expression from {l _i } supplied from the basic cycle path candidate configuration unit, and selects it as a basic cycle book. To the selection unit 60, the lag corresponding to the (N-1) th sub-frame

[0144]

[0145] So I deleted it.

FIG. 10 shows one embodiment of the basic cycle preliminary selecting section in the ninth invention. Basic cycle preliminary selection unit 94
The basic cycle extraction unit 944 of 0 calculates the cumulative prediction gain from {l _i } input from the basic cycle path candidate configuration unit 43, and selects a desired lag set from the threshold memory 945. The threshold value gn _th is input and the cumulative value is calculated only for the prediction gain exceeding the threshold value. That is,

[0147]

The set of lags l _imax output to the basic period main selection unit 60 maximizes the above equation.

FIG. 11 is a block diagram showing another embodiment of the basic cycle preliminary selecting section of the ninth invention. In the integer lag prediction gain calculation unit 1042 of this embodiment, the prediction gains described so far are described.

[0150]

In the calculation of, the lag t is limited to an integer value. Based on the prediction gain of the integer lag calculated by the integer prediction gain calculation unit 1042, the basic cycle path candidate configuration unit 1043, for example, first gives a cycle tmax _nsub that gives the maximum prediction gain and its maximum value.

[0152]

A set of lags that satisfies the following relationship is obtained. The {l _i } thus obtained is output to the decimal lag prediction gain calculation unit 1045 and the candidate basic period extraction unit 1044. The fractional lag prediction gain calculation unit calculates a prediction gain of a fractional lag in the vicinity of a predetermined lag set {l _i } supplied from the basic cycle path candidate configuration unit 1043. The prediction gain corresponding to the decimal lag t _frac is, for example, t _frax = t _i + t _f / D (D is an integer value)
In case of, it can be calculated as follows.

[0154]

Here, h () is the impulse response of the polyphase filter described in Reference 4. In the candidate basic cycle extraction unit 1044, based on {l _i } supplied from the basic cycle path candidate configuration unit and the decimal lag prediction gain near {l _i } supplied from the decimal lag prediction gain calculation unit 1045. The maximum cumulative prediction gain of the following equation from the neighborhood of {l _i }

[0156]

Here, W ^b is a predetermined constant.

In the above embodiments, the configuration in which all subframes are differentially encoded has been described. However, a configuration in which the subframes are divided into several strings and the differential encoding is performed for each set is adopted. You can also This configuration is effective when the fundamental period of the input audio signal in the frame is largely changed.

[0159]

As described above, according to the present invention, in the configuration in which the basic period is differentially encoded, the basic period preliminary selection unit performs the optimum tracking by evaluating the distortion over the entire frame, and the basic period main selection unit performs the tracking. Provided is a basic cycle coding device capable of coding an optimum basic cycle for a whole frame at a low bit rate by obtaining a basic cycle from the vicinity of a candidate basic cycle supplied from the basic cycle preliminary selection unit and performing differential coding. The effect is that you can do it.

[Brief description of drawings]

FIG. 1 is a block diagram of a speech coding apparatus including an embodiment of the present invention.

FIG. 2 is a block diagram of a basic cycle preliminary selection unit in the first invention.

FIG. 3 is a block diagram of a basic cycle preliminary selection unit in the second invention.

FIG. 4 is a block diagram of a basic period main selection unit in a third invention.

FIG. 5 is a block diagram of a basic cycle preliminary selection section and a basic cycle main selection section in a fourth invention.

FIG. 6 is a block diagram of a basic cycle preliminary selection section and a basic cycle main selection section in a fifth invention.

FIG. 7 is a block diagram of a basic period preliminary selection unit and a basic period main selection unit according to a sixth aspect of the present invention.

FIG. 8 is a block diagram of a basic cycle preliminary selection unit in a seventh invention.

FIG. 9 is a block diagram of a basic cycle preliminary selection unit in an eighth invention.

FIG. 10 is a block diagram of a basic cycle preliminary selection unit in a ninth invention.

FIG. 11 is a block diagram of a basic cycle preliminary selection unit in the ninth invention.

[Explanation of symbols]

10 Input Terminal 11 Basic Period Output Terminal 12 Sound Source Codebook Index Output Terminal 13 Gain Codebook Index Output Terminal 14 Quantization LPC Index Output Terminal 20 Frame Buffer 30 LPC Analysis Section 31 LPC Memory 32 LPC Interpolation Section 40, 440, 540, 640 , 740, 840, 94
0, 1040 Basic cycle preliminary selection section 41, 841 Weighting section 42, 842 Prediction gain calculation section 43, 243, 543, 643, 843, 1043 Basic cycle path candidate configuration section 44, 744, 844, 944, 1044 Candidate basic cycle extraction Part 50 LPC quantizer 51 Quantized LPC memory 52 Quantized LPC interpolator 60, 460, 560, 660 Basic period main selection unit 70 Excited sound source search unit 80 Excited sound source memory 90 Gain search unit 242 Maximum predicted gain extraction unit 361 Distance Calculation unit 362, 462, 562, 662 Candidate lag configuration unit 363 Basic period lag selection unit 364 Excitation sound source configuration unit 365 Basic period lag memory 501 Candidate range table 601 Frame mode identification unit 601 Class index output terminal 820 2 Frame buffer 845 Candidate basic circumference Memory 945 threshold memory 1042 integer lag prediction gain calculator 1045 fractional lag prediction gain calculator

Claims (9)

[Claims] 1. A speech coding method in which a discrete speech signal is represented by a parameter representing a spectrum envelope, a parameter representing a fundamental period, a parameter representing an excitation source, and a parameter for performing amplitude control, and divided into frames at regular intervals. When a discrete speech signal is input and the basic period parameter is encoded for each subframe obtained by dividing the frame into a plurality of sections, some of the basic period parameters of the subframe are set as basic period parameters of the subframe determined in the past. In the basic cycle coding apparatus for coding with the difference of, a basic cycle preliminary selecting section that outputs a set of candidate basic cycles corresponding to each subframe of the frame based on the input speech signal, and the basic cycle preliminary selection Of the set of candidate basic periods output from the sub-frame Basic cycle main selection unit that determines a basic cycle from the input basic speech signal of the subframe, the spectral phrase parameter of the subframe, and the parameter representing the excitation sound source determined in the past from the neighboring basic cycle parameters in the range In the process in which the basic period preliminary selection unit obtains a candidate basic period of a subframe over the entire frame by the basic period preliminary selection unit, a plurality of basic units for each subframe are obtained using the input audio signal. A period parameter candidate and a prediction gain associated with the candidate are defined, and a difference in the basic parameter between adjacent subframes based on each of the defined basic period parameter candidates is within a predetermined range. , And the prediction gains associated with them from the set of fundamental period parameters over the constructed frame. Basic cycle encoding apparatus and extracts the candidate base period and. 2. A subframe that gives a maximum prediction gain using the input speech signal in the process of the basic period preliminary selection unit obtaining candidate basic period parameters of a subframe over the entire frame in the basic period preliminary selection unit. And
In the fixed subframe, a basic period parameter candidate and a prediction gain associated with the candidate are determined based on the input speech signal, and a basic parameter between adjacent subframes based on each of the determined basic period parameter candidates. A basic period parameter set whose difference is within a predetermined range, and extracting a candidate basic period from the set of basic period parameters over the configured frame based on prediction gains associated with them. The basic period coder according to claim 1, wherein: 3. The basic cycle main selection unit configures a set of neighboring basic cycle parameters within a predetermined range from a basic cycle determined in a past subframe, and selects from the configured neighboring basic cycle parameters. The basic period coding apparatus according to claim 1, wherein the basic period is determined using an input signal of the subframe, a spectrum envelope parameter of the subframe, and a parameter representing an excitation source determined in the past. 4. The basic period coding apparatus, wherein the basic period preliminary selecting section outputs a plurality of candidate basic periods. 5. The basic period encoding apparatus according to claim 1, wherein the number of pits assigned to the basic period to be differentially encoded is different for each subframe. 6. A mode identifying section for dividing the frame or the subframe into predetermined classes based on a prediction gain for each subframe calculated from the input speech signal. 2. The basic period coding apparatus according to claim 1, wherein the number of pits assigned to the basic period to be differentially encoded is made different for each subframe based on the output class information. 7. The basic cycle preliminary selecting unit extracts a candidate basic cycle from a set of a plurality of lags configured such that a difference in basic parameter between adjacent subframes is within a predetermined range. , The prediction gain associated with each lag is used to narrow down the set of basic parameters into a smaller number of sets, and the highest basic period is extracted from the set of narrowed down basic parameters. 1. The basic period coder according to 1. 8. A predictive gain determined by a preceding frame, an input speech signal of the next frame, and an input speech signal of a next frame in a process in which the basic cycle preliminary selecting section configures a candidate basic cycle of a subframe by the basic cycle preliminary selecting section Based on the above, the difference in the basic parameter between adjacent subframes is predetermined over a plurality of frames including a part of the subframe of the previous frame, a subframe of the frame and a part of the subframe of the next frame. A basic period parameter set within a range is formed, and a candidate basic period is extracted from a set of basic period parameters formed over the plurality of frames, based on prediction gains associated with them. Item 1. The basic period encoding device according to item 1. 9. The basic cycle preliminary selection unit configures a plurality of sets of basic cycle parameters within a predetermined range from basic parameter candidates of adjacent subframes, and sets the configured basic cycle. When determining a candidate basic cycle from a set of parameters, the basic cycle is extracted using a basic cycle parameter that gives a prediction gain exceeding a predetermined threshold value among the elements of the set of basic parameters configured as described above. The basic period encoding device according to claim 1.